Blog

Ethical approval

The main trial (PregnAnZI-2) was approved by the Gambia Government-MRCG Joint Ethics Committee, the Comité d’Ethique pour la Recherche en Santé, the Ministry of Health of Burkina Faso, and the London School of Hygiene and Tropical Medicine Ethics Committee. In addition, this post hoc study was approved by the Gambia Government-MRCG Joint Ethics Committee and the London School of Hygiene and Tropical Medicine Ethics Committee. Study women signed informed consent for the trial during their antenatal visits and samples collected during the 4-months follow-up. An additional consent was sought from mothers of the study children for the 3-year survey.

Study design

PregnAnZI-2 (ClinicalTrials.org NCT03199547) was a phase-III, double-blind, randomized, placebo-controlled trial in which 11,983 women from The Gambia and Burkina Faso were randomized to receive a single dose of 2 g of oral azithromycin or placebo (ratio 1:1) during labor. The median time from administration of azithromycin to delivery was 1.6 h (IQR 0.5–4.1)⁹. Details of inclusion and exclusion criteria are available in the study protocol published elsewhere³⁶. The primary objective of the trial was to evaluate the impact of the intervention on neonatal sepsis and death⁹. The trial showed no effect on the neonatal sepsis or death though other infections, including skin infections, were reduced in the azithromycin arm⁹. Also, the intervention reduced maternal infections including mastitis⁹.

Here only participants from sites in The Gambia were included, where the trial was conducted in two health facilities located in the coastal region of the country. A subset of 253 mother-baby pairs were recruited in a bacterial carriage sub-study. These were randomly selected from study children born between January 2019 and March 2020³⁶. All the children in the carriage sub-study were vaginally delivered.

Selection of children for mycobiota analysis

The first 102 children recruited into the bacterial carriage sub-study in the Gambia were selected. For these children, an additional follow up was conducted at the age of 3 years, during which rectal swabs were collected from 72 of the 102 children initially selected plus an additional 25 children from the main carriage cohort. Details of the sample selection process is illustrated in Fig. 5. A total of 126 children were included in this mycobiota study. Fifty-seven of these children were born to mothers who received the active intervention (2 g azithromycin) and 69 were born to mothers who received placebo.

Collection of rectal swabs

During the trial, rectal swabs (RS) were collected with FLOQSwabs (COPAN, REF:519CS01) from the children included in the bacterial carriage sub-study. The first swabs were collected shortly after delivery (day 0). After post-delivery hospital discharge, active-follow up with home-visits occurred at day 6, day 28 and 4 months. RS were collected from the children during these follow-up visits. All the swabs were stored in skim milk-tryptone-glucose-glycerin (STGG) transport medium without preservative and transported in temperature-monitored (2–8 °C) cooler boxes with ice packs to the laboratory within 8 h. Upon reception, the samples were homogenized and stored at −80 °C for later processing.

Rectal swabs used for mycobiota analysis

A total of 467 RS from the children (214 azithromycin, 253 placebo) were processed. Sample numbers per time-point ranged from 40 to 46 for azithromycin arm and from 48 to 53 for placebo arm (Supplementary Table 1). As per best practice for microbiome studies, field controls were collected during the main study (PregnAnZI-2 trial) and during the year 3 follow up of children included in the microbiome studies. A vial of STGG with a plain sample collection swab immersed and exposed for approximately 1 min to the environment where participants were sampled, was used as a field control. To avoid contamination with fecal matter from the samples, the field controls were collected before the rectal swabs. A total of 109 field controls were collected, of which 12 were collected at the hospital sites during the main trial follow ups and 97 were collected from individual participant households during the year 3 follow up.

DNA extraction and quality control

We extracted DNA from the RS using the DNeasy® PowerLyzer® PowerSoil® Kit from Qiagen (Qiagen, Germany) with slight modifications as follows. A total of 100 µl of homogenized RS in 1 ml of STGG transport medium was used as input and two rounds of 10 cycles of beat beating at 2500 RPM for 30 s was applied with 10 min break between rounds. We followed the manufacturers protocol for the rest of the extraction steps and eluted in 100 µl of elusion buffer from Qiagen. Blank extraction and field controls were included and taken through all downstream analyses. Plain sterile swabs in sample storage vials with STGG exposed to the environment at the field site were included as field controls. The DNA extracts were quantified with a Qubit 4·0 Fluorometer (Invitrogen/Thermo Fisher Scientific) using the double stranded DNA high-sensitive kit.

ITS2 library preparation and sequencing

ITS2 amplicons were generated using the following primers (forward: GTGAATCATCGARTCTTTGAAC, reverse: TATGCTTAAGTTCAGCGGGTA) published elsewhere³⁷, and the Q5 Hi-Fi 2X master mix from New England Biolabs (NEB). The PCR reaction was set up with 12.5 µl of master mix, 1 µl each of forward and reverse primers at 10 µM concentration, 5 µl of template, and 5.5 µl of molecular grade water. The following thermocycling conditions were used. Initial denaturation at 98 °C for 30 s, followed by 35 cycles of denaturation at 98 °C for 20 s, annealing at 51 °C for 30 s, and extension at 72 °C for 20 s. This was followed by a final extension at 72 °C for 10 min. The expected amplicon size was ~350 bp. Post PCR DNA concentrations were quantified as described earlier. The PCR products where then normalized to 50 ng (~200 fmol) in 12.5 µl for library preparation. Nanopore libraries were made using the ligation sequencing kit (SQK-LSK109) with the native barcoding expansion 96 (EXP-NBD196) using manufacturers protocol. Six pools of 96 libraries and 1 pool of 39 libraries were generated including rectal samples, field controls, extraction reagent blanks and PCR reagent blanks. Each pool was sequenced using an R9 MinION flow cell (FLO-MIN106D) on a GridION using quality filters of 200 bp minimum read length and minimum quality score of Q9. Between 4 million to 8 million reads were generated for each of the 96 library pools, and about 400,000 reads were generated for the pool of 39 libraries, which were mostly field controls.

Analysis of ITS2 sequences

The sequence data were analyzed using NanoCLUST³⁸, a pipeline designed for analyzing 16S rRNA gene amplicon data generated on the Oxford Nanopore Technology (ONT) sequencing platform. The pipeline was optimized for analyzing ITS2 data generated on the ONT sequencing platform. The pipeline takes in raw FASTQ files and performs trimming and filtering using fastp³⁹. Reads were filtered by minimum Q-score of 9 and read lengths between 200 bp and 500 bp. It then generates k-mer frequencies and performs clustering using UMAP-HDBSCAN^40,41. We subsampled up to 5000 reads per sample for clustering setting minimum cluster size at 50 and minimum distance to define a cluster at 0.1. Sequences in each cluster are then corrected using Canu assembler⁴². This is followed by the selection of draft representative sequences for each cluster, which are subsequently polished using Racon⁴³, and Medaka⁴⁴. The polished consensus sequences are then classified using Kraken2⁴⁵, with a subset of the standard Kraken2 database (April 2024 version) containing only fungal and human sequences.

Mycobiota community analysis

Classification data generated with NanoCLUST were imported into R (version 4.4.0) and RStudio (version 2024.04.1 + 748)⁴⁶, for further microbial community analysis. The Tidyverse (version 2.0.0), and phyloseq (version 1.48.0)⁴⁷, packages were mainly used for data handling and organization. First, taxon counts were normalized by total sum scaling. We then inspected the initial profiles of the samples and controls to see the relative abundances of human and unclassified reads as well as possible contaminants (Supplementary Fig. 1). Most of the reads in the PCR and extraction blanks were unclassified, and the top two taxa in the rectal swabs were unclassified and human (Supplementary Fig. 1). We filtered out all human and unclassified reads. The filtered dataset was transformed into a feature table and used to create a phyloseq object together with the sample metadata. Using decontam (version 1.20.0)⁴⁸, we identified contaminants in the dataset using the isContaminant function in the combined mode setting a probability cutoff of 0.2 for calling a contaminant. A total of 18 taxa (Supplementary Table 2) were identified as contaminants and filtered from the dataset. We re-inspected the profiles of the samples and controls again to see how they have been impacted by the filtering. Most of the potential contaminants were removed, though there still remained a couple of other potential contaminants (Aspergillus fumigatus and Aspergillus puulaauensis) which were subsequently filtered out (Supplementary Fig. 2). Details on the sample numbers that were available before and after all the filters were applied are shown in Supplementary Table 1. The number of reads in the samples and controls before and after filtering of unclassified, human, and contaminants are shown in Supplementary Fig. 3. Samples with only 1 or 2 species (not read counts) remaining post-filter were excluded from analyses of community composition, clustering, and differential abundance to avoid outlier effects.

Before measuring the effects of the intervention on the mycobiota, we first assessed the trial arms for comparability at baseline. Variables that had different representation between the trial arms were controlled for during statistical analysis.

We measured diversity within individual samples (alpha diversity) using Shannon index and species richness. To understand the effects of potential factors that influence microbiome diversity, we assessed the effects of treatment, ethnicity, season of sampling, parity, and sex (sex was determined at birth) among others as covariates on Shannon diversity and species richness using a linear mixed effects model with a random effect on subject. Based on this model we examined interaction between treatment and season measuring variation of Shannon diversity between trial arms at each time-point by season using Wilcoxon rank sum test. The season stratified analysis did not include samples collected at the age of 3 years because all the samples at this time point were collected during the wet season. We also examined interaction between treatment and parity in a similar analysis. Finally, we assessed the effect of age on Shannon diversity using a linear mixed effects model, comparing day 0 to each of the subsequent time-points while controlling for treatment, season of sampling, ethnicity, parity and sex and adding individual variations as random effects. p values were adjusted for repeated testing by Dunnett’s test.

Overall community composition was measured with Bray–Curtis index using the vegdist function in vegan (version 2.6.4)⁴⁹. Variance in community composition was compared between groups by permutational multivariate analysis of variance (PERMANOVA) using the adonis function in vegan. We assessed the effects of age, season of sampling, ethnicity, parity, and sex on overall community composition within each trial arm. For age comparisons, we compared variance across all time-points and then between day 0 and each of the subsequent time-points. We then compared variance in community composition by treatment by comparing trial arms at each time-point. As we did not find any significant variation in community composition by sampling season, parity, ethnicity, or sex, no stratified analysis was done for these variables.

To further explore the structure of the gut fungal community in our cohort, we employed an unsupervised clustering using Dirichlet’s Multinomial Mixtures model implemented in mothur (version 1.44.0)⁵⁰, to group samples into clusters based on their community types. We used logistic regression to examine association between fungal clusters (grouped as dominant community type which we called fungCluster_3 vs others) and covariates including treatment, age, sampling season, and parity. We visualized the profiles of the clusters using a heatmap to see which taxa were driving the separation of samples and summarized the frequencies of the clusters per time-point for each trial arm. To understand the dynamics of changes in fungal community types, we assessed the transition of individual children through the different clusters over time.

We assessed variations in taxon abundance using linear models implemented in MaAsLin2 (version 1.18.0)⁵¹. First, we assessed temporal variations in taxon abundance between the trial arms at each time-point. We then assessed differences in taxon abundance between trial arms by season of sampling. Finally, we assessed overall abundance for each taxon by treatment, age, season of sampling, parity, ethnicity, sex, and bacterial cluster, while adjusting for random effects on subject. Only taxa with minimum abundance and variance of at least 0.01% were included in differential abundance analysis. p values were adjusted for multiple testing and an FDR cutoff of 0.2 was used for calling significance. We used bacterial clusters identified in these children from our previous study²², where we characterized gut microbiota development to identify potential associations of the bacterial community types with the abundance of individual fungal taxa. Bacterial clusters identified in the children at day 28 were used as this was the point at which we saw marked differences in the representation of bacterial community types between the trial arms²². Data on the bacterial clusters assigned to individual samples is provided in Supplementary Data 1⁵².

Statistics and reproducibility

Sample size for each trial arm at baseline is shown in Table 1. Demographic comparability of the trial arms at baseline was assessed using various tests. For categorical variables (ethnicity, sex, season, breastfeeding mode) except for recent sickness and recent antibiotic consumption, Chi-square test was applied. For recent sickness and recent antibiotic consumption, Fisher’s exact test was applied. For continuous data (maternal age, birthweight, breastfeeding duration), t-test was applied.

The sample size was originally calculated for gut microbiota analysis. We chose to work with the same set of samples because we intended to link our findings with the results we obtained from the gut microbiota analysis. The sample size calculation was based on power to detect at least 10% difference in the top 10 operational taxonomic units (OTUs) and 20% difference in the next 10 OTUs in the gut microbiota using a sample size and power calculation tool for case-control microbiome study design developed by Mattiello et al. ⁵³. We used the gut microbiome dataset from the human microbiome study embedded in the tool. With a sample size range of 30–70 per group, we estimated power by Monte Carlo simulations with 100 replications using the top 50 OTUs from the dataset. A sample size of 45 per group at each time-point had over 90% power to detect these differences. Though we expect higher interpersonal variations with the gut mycobiota than the gut microbiota, our analysis factored this by exclusion of outliers in comparisons of community composition and differential taxon abundance. The remaining samples used in each of these analyses are shown in the respective results.

Multivariate analysis of factors that influenced alpha diversity was done using a linear mixed effects model with random effects on individuals. Temporal variations in alpha-diversity between trial arms was assessed at each time-point by Wilcoxon Rank Sum test. Changes in Shannon diversity over time was assessed using a linear mixed effects model including ethnicity, season, parity, and sex as covariates in addition to the trial treatment while controlling for random variations among individuals. p values were adjusted by Dunnett’s test.

Beta-diversity was compared between trial arms at each time-point by permutational multivariate analysis of variance (PERMANOVA) using the adonis function in vegan⁴⁹. Beta-diversity was also compared by age across timepoints using the adonis function with permutations restricted within individuals.

Differential taxon abundance was assessed using generalized linear models implemented in MaASLin2⁵¹. Temporal differences between trial arms were assessed at each time. Variations by age and other covariates were assessed with the aggregated data.

Analysis of factors that drive mycobiota community-types was done by logistic regression using a generalized linear model.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Natural processes on land and in the ocean already remove carbon dioxide from the air, but they do not work in isolation. A new paper argues that these processes must be seen as one connected system, a weathering continuum that stretches from high terrain to the deepest seabed and controls how much CO2 is locked away over time.

This perspective matters because many climate ideas aim to speed up weathering to pull more CO2 from the atmosphere. If the pieces of the chain are tightly linked, pushing on one part can tug on another in ways that help or hurt the total carbon removed.

Why linking land and ocean matters

Dr. Gerrit Trapp-Müller of the Georgia Institute of Technology led the study, which was completed during his time at Utrecht University.

The team brings reactions in soils, rivers, coasts, and marine sediments into one model so we can view the whole system, not just scattered parts.

In this framework, the direction and size of a local flux depend on the material’s origin, how it has been transported and altered, and the surrounding conditions. That is a practical shift, because it changes how we should place, size, and verify interventions meant to boost carbon removal.

How silicate weathering stores CO2

At the center is silicate weathering – the set of reactions where acidic water breaks down silicate minerals, releasing cations and bicarbonate that ultimately store carbon as carbonate minerals.

Over geologic time, silicate weathering acts as a negative feedback that steadies climate by consuming more CO2 as temperatures rise.

This thermostat effect is real but slow on human timescales. It scales with factors such as temperature, runoff, rock type, and how quickly fresh mineral surfaces are exposed.

Oceans can release CO2

There is a twist called reverse weathering. In marine sediments, the formation of authigenic clays can consume alkalinity, release acidity, and shift carbon back toward the atmosphere.

When reactions that consume alkalinity dominate parts of the chain, the net sink can shrink or even flip. That means the ocean side of the system can throttle what the land side achieves.

Not all settings pull equal weight. Recent work highlights deltas and beaches as marine weathering hotspots that can shape the balance between forward and reverse reactions along the transport path from rivers to the shelf and beyond.

Studies in delta muds, using both field data and models, show that where the sediment comes from and how it was deposited determine whether chemical reactions add or remove alkalinity – and by how much – at different depths and over time. These factors decide whether the area is a strong sink or only a weak one.

How much CO2 weathering removes

So how much carbon does chemical weathering handle today. A global analysis finds total CO2 consumption by chemical weathering near 237 million metric tons of carbon per year, with silicates accounting for roughly 63 percent of that budget, on the order of 149 million tons of carbon annually.

Within the silicate share, basaltic terrains have an outsized impact, contributing on the order of 25 to 35 percent of the silicate weathering CO2 sink because they weather quickly where runoff is high.

That concentration of flux in a small fraction of the landscape explains why placement of interventions matters so much.

How we could boost weathering

Analyses suggest that spreading finely crushed silicate rock on croplands could remove on the order of 0.5 to 2.0 billion tons of CO2 per year, comparable to other land-based options if done at scale and with the right logistics.

This is a large range, and it depends on climate, soils, farm practices, and supply chains.

Recent evaluations also warn that removal efficiency varies widely and can be overestimated if you ignore side fluxes, carbonate dynamics, or downstream processes in the ocean that offset gains on land. Measuring the whole chain is not optional, it is the only way to know the real net.

The future of weathering research

“The main conclusion from our work is that the various CO2 fluxes on land and in the ocean are very closely linked. This governs the efficiency of the removal of CO2 from the atmosphere,” said Trapp-Müller.

Treating weathering as one system forces three practical habits. Site selection should consider downstream coupling, monitoring must track products beyond the field edge, and models need to include both forward and reverse reactions across environments.

Future studies may map which river basins, coastlines, and sedimentary settings deliver the biggest net gains once ocean processes are included. That includes quantifying how fast material moves along the conveyor from hills to seabed and how reaction balances change en route.

The new perspective also invites better stress tests for proposed projects, from agricultural basalt applications to coastal alkalinity additions, against the actual carbon accounting of the full chain. 

The study is published in the journal Nature Geoscience.

—–

Like what you read? Subscribe to our newsletter for engaging articles, exclusive content, and the latest updates. 

Check us out on EarthSnap, a free app brought to you by Eric Ralls and Earth.com.

—–

The catalogue, which identifies five times as many canyons as previous studies had, was produced by the researchers David Amblàs, of the Consolidated Research Group on Marine Geosciences at the Faculty of Earth Sciences of the University of Barcelona, and Riccardo Arosio, of the Marine Geosciences Research Group at University College Cork. Their article shows that Antarctic submarine canyons may have a more significant impact than previously thought on ocean circulation, ice-shelf thinning and global climate change, especially in vulnerable areas such as the Amundsen Sea and parts of East Antarctica.

Submarine canyons: the differences between East and West Antarctica

The submarine canyons that form valleys carved into the seafloor play a decisive role in ocean dynamics: they transport sediments and nutrients from the coast to deeper areas, they connect shallow and deep waters and they create habitats rich in biodiversity. Scientists have identified some 10,000 submarine canyons worldwide, but because only 27% of the Earth’s seafloor has been mapped in high resolution the real total is likely to be higher. And despite their ecological, oceanographic, and geological value, submarine canyons remain underexplored, especially in polar regions.

“Like those in the Arctic, Antarctic submarine canyons resemble canyons in other parts of the world,” explains David Amblàs. “But they tend to be larger and deeper because of the prolonged action of polar ice and the immense volumes of sediment transported by glaciers to the continental shelf.” Moreover, the Antarctic canyons are mainly formed by turbidity currents, which carry suspended sediments downslope at high speed, eroding the valleys they flow through. In Antarctica, the steep slopes of the submarine terrain combined with the abundance of glacial sediments amplifies the effects of these currents and contributes to the formation of large canyons.

The new study by Amblàs and Arosio is based on Version 2 of the International Bathymetric Chart of the Southern Ocean (IBCSO v2), the most complete and detailed map of the seafloor in this region. It uses new high-resolution bathymetric data and a semi-automated method for identifying and analysing canyons that was developed by the authors. In total, it describes 15 morphometric parameters that reveal striking differences between canyons in East and West Antarctica.

“Some of the submarine canyons we analyzed reach depths of over 4,000 meters,” explained David Amblàs. “The most spectacular of these are in East Antarctica, which is characterized by complex, branching canyon systems. The systems often begin with multiple canyon heads near the edge of the continental shelf and converge into a single main channel that descends into the deep ocean, crossing the sharp, steep gradients of the continental slope.”

Riccardo Arosio noted that “It was particularly interesting to see the differences between canyons in the two major Antarctic regions, as this hadn’t been described before. East Antarctic canyons are more complex and branched, often forming extensive canyon-channel systems with typical U-shaped cross sections. This suggests prolonged development under sustained glacial activity and a greater influence of both erosional and depositional sedimentary processes. In contrast, West Antarctic canyons are shorter and steeper, characterized by V-shaped cross sections.”

According to David Amblàs, this morphological difference supports the idea that the East Antarctica Ice Sheet originated earlier and has experienced a more prolonged development. “This had been suggested by sedimentary record studies,” Amblàs said, “but it hadn’t yet been described in large-scale seafloor geomorphology.”

About the research, Riccardo Arosio also explained that “Thanks to the high resolution of the new bathymetric database — 500 meters per pixel compared to the 1-2 kilometres per pixel of previous maps — we could apply semi-automated techniques more reliably to identify, profile and analyse submarine canyons. The strength of the study lies in its combination of various techniques that were already used in previous work but that are now integrated into a robust and systematic protocol. We also developed a GIS software script that allows us to calculate a wide range of canyon-specific morphometric parameters in just a few clicks.”

Submarine canyons and climate change

As well as being spectacular geographic accidents, the Antarctic canyons also facilitate water exchange between the deep ocean and the continental shelf, allowing cold, dense water formed near ice shelves to flow into the deep ocean and form what is known as Antarctic Bottom Water, which plays a fundamental role in ocean circulation and global climate.

Additionally, these canyons channel warmer waters such as Circumpolar Deep Water from the open sea toward the coastline. This process is one of the main mechanisms that drives the basal melting and thinning of floating ice shelves, which are themselves critical for maintaining the stability of Antarctica’s interior glaciers. And as Amblàs and Arosio have explained, when the shelves weaken or collapse, continental ice flows more rapidly into the sea and directly contributes to the rise in global sea level.

Amblàs and Arosio’s study also highlights the fact that current ocean circulation models like those used by the Intergovernmental Panel on Climate Change do not accurately reproduce the physical processes that occur at local scales between water masses and complex topographies like canyons. These processes, which include current channeling, vertical mixing and deep-water ventilation, are essential for the formation and transformation of cold, dense water masses like Antarctic Bottom Water. Omitting these local mechanisms limits the ability that models have to predict changes in ocean and climate dynamics.

As the two researchers conclude, “That’s why we must continue to gather high-resolution bathymetric data in unmapped areas that will surely reveal new canyons, collect observational data both in situ and via remote sensors and keep improving our climate models to better represent these processes and increase the reliability of projections on climate change impacts.”